[surface_matching] Add binary save/load for PPF3DDetector

[v5a]

Summary

This PR implements saveModel and loadModel methods for PPF3DDetector to enable saving trained models to disk and reloading them later. This addresses the long‑standing missing functionality reported in #3353.

Changes

• Added saveModel(): writes all internal state (sampled point cloud, PPF matrix, hash nodes, hash table) to a custom binary file.
• Added loadModel(): reads back the binary file and reconstructs the detector state exactly.
• A custom binary format is used instead of FileStorage because XML/YAML serialization is too slow for the large datasets typically processed by this module.

Testing

• Trained a model with the example model parasaurolophus_6700.ply.
• Saved the model to trained_model.bin and loaded it into a new detector instance.
• Matched both detectors against the scene rs1_normals.ply.
• Both detectors found exactly the same 20 poses; the Frobenius norm of the first pose difference is 0.0, confirming perfect consistency.

Related Issue

Closes #3353

[v5a]

Hi @asmorkalov,
The two failing checks (CUDA and FastCV) appear to be unrelated to the changes in this PR. Could you please take a look? All other platforms passed successfully. Thanks!